Proof
How we know the model works.
Every number below comes from an out-of-time test: the model that scores a quarter was trained only on quarters before it, exactly as it would have been in production.
The method
Leak-free, quarter by quarter.
We split history into quarters. For each test quarter the model is trained on data available before that quarter began, then asked to rank every single-family home in Cuyahoga County. Twelve months later we check which homes actually listed.
Top-decile lift is the share of sellers among the model's top 10% divided by the share among all homes. Precision at the top 1% is the share of the model's 1% most confident homes that listed within the year.
The table shows the deployed survival model on five consecutive test quarters.
| Test quarter | County base rate | Top-decile lift | Precision at top 1% |
|---|---|---|---|
| 2021-04-01 | 6.1% | 3.31x | 33.2% |
| 2021-07-01 | 6.0% | 3.37x | 32.8% |
| 2021-10-01 | 5.7% | 3.44x | 32.6% |
| 2022-01-01 | 5.3% | 3.51x | 30.3% |
| 2022-04-01 | 5.1% | 3.51x | 30.7% |
What it means
A call list that pays off three times as often.
If an agent works the top 10% of the model's list, roughly 1 in 5 of those homes lists within the year, against about 1 in 18 countywide. Work the top 1% and about 1 in 3 lists.
Those are the homes to call first, and the reasons behind each score tell you what to say.
Caveats
What the numbers do not say.
- Results are for Cuyahoga County, Ohio, where the model has the deepest history. New markets are validated the same way before agents see a score.
- Base rates move with the market; lift is measured against the base rate of the same quarter.
- A locked live cohort from July 2026 is still maturing and is not included above.
- No model predicts an individual sale. HiTouch ranks; you still have to call.